3D models acquired from geometry scanners are usually noisy and irregular. Generally, the acquired 3D data needs to be de-noised and the approximated surface to be smoothed. When a de-noising filter is applied to smooth 3D models, the performance generally depends on the size of the filter. If the size of the filter is too small, noise may not be eliminated sufficiently or more de-noising iterations may be required. If the size is too large, fine details may be oversmoothed. It remains a challenge to remove noise while preserving the fine details of 3D models.